Always-on AI, built into your devices.
Local language and voice models work alongside you as tasks unfold and plans change, with frontier models available when needed.
A foundation model company.
Beyond Intelligence.
To make intelligence a lasting human capability.
Introduction
Distributing intelligenceto all of humanity.
Our view
Three changes to how AI works.
Make intelligence a reusable resource.
Work completed on device has zero inference cost. Developers can build new products around continuous assistance and frequent use.
Let people shape intelligence.
Our models give people and organizations a foundation to shape AI around their own knowledge and purposes, opening possibilities we cannot anticipate.
SADAIR Research
We believe AI’s next chapter will be shaped by discoveries made wherever intelligence is used. Our research connects local intelligence with the frontier to expand the possibilities of human invention.
Read the research- The Second History of IntelligenceHow intelligence becomes lasting human capability under personal authority.
- Building full-duplex intelligenceKeeping a live conversation on the device while reasoning and tools work in the background.
- Continual learning on personal devicesHow a shared model becomes each person’s own through local weight updates.
Our modelsResearch in development
What a personal language model does on your phoneRead the PLM section
You
PLM
Reasoned on the device from your itinerary and the local time.
65tok/s
PLM 1.8B, iPhone 17 Pro
0.67s
First token
$0
Cost on the device
BenchmarksSADAIROthers and references
Decode speed on iPhone 17 Pro
Tokens per second, higher is better
PLM 1.8B65.0
PLM 3B43.0
Qwen3.5-4B33.9
MATH-500, pass@1
Accuracy, percent
PLM 8B-A3B86%
Qwen3.5-4B80.8%
Gemma 4 E4B65.0%
Granite 4.0 H-Tiny59.2%
Measurement notes
Personal Language Model
Phone performance uses Q4_K_M on iPhone 17 Pro with 1,024 input tokens, 256 output tokens, and batch size one. Results exclude model loading and initial compilation. Capability scores use BF16 checkpoints evaluated by SADAIR under the same protocol.
Results and methods in the researchIn practice
Intelligence, in practice.
See our language and voice models work together in one continuous conversation.
On-device demo
KRORAINA 2 carries the conversation, PLM draws on personal context, and Smart Router brings in the models and services the moment requires.
Watch the demoOur partners
Selected collaborations
Explore what we could build together.
Work with us on device integration, custom models, and new research.
Join SADAIRSan Francisco or remote
Build for what people will invent.
We are looking for people who can move between first principles and finished work.
Explore careers
